This thesis is based on the critical analysis of publicly available policy documents relating to international students. Carol Bacchi’s WPR approach provides a bridging framework to mediate between Foucault’s high-level theory (1972) and the policy texts themselves. Without sacrificing the complexity of a discursive analysis, her framework of questions provides a structure on which to hang a potentially nebulous enquiry. However, Bacchi’s approach is typically employed in relatively coherent and well-defined policy areas, often with single texts or single genres of policy document. This study operates within a diffuse policy with multiple fields and multiple genres. As Knight (2004, p.17) suggests, on a national policy level international higher education policy includes “(e)ducation and other national-level policies relating to international dimension of higher education; other policy sectors include cultural, scientific, immigration, trade, employment, and culture.” Therefore, I have conducted a preliminary qualitative analysis to make sense of the policy field before applying Bacchi’s framework.
The qualitative analysis used NVivo software to facilitate an inductive thematic coding. Relationships between these themes were organised around rationales for or against recruiting international students. These rationales constituted the starting point for Bacchi’s (2009; 2012) “what is the problem represented to be” (WPR) analysis. This framework was then applied to identify how international students are
discursively represented in policy.
This chapter explains firstly how documents were identified and selected, secondly how documents were coded and how NVivo was used to supplement the process, and finally how the qualitative analysis related to the WPR analysis.
64
The texts
In keeping with the discursive analytical approach described in Chapter 3, a text- based method was used. Texts are understood here as snapshots of policy discourses, selectively constructed and socially produced, such that the choices around language and content reveal ideologies.
Selection
Policy documents were included if they:
1. Were published between 1999 and 2013.
2. Had international students as their main object or potentially impact them; 3. Was published by a central government agency or centrally funded quasi-
governmental organisation (classified as primary), or was referenced frequently in such documents and, therefore, understood as influential in policy formation (classified as secondary)
4. Were publicly accessible or available.
Below, I expand on each of these criteria.
1. The documents were selected to illustrate the period between 1999 when the PMI was launched, and 2013 when the Coalition International Education Strategy (IES) was launched, in order to identify significant differences in policy. In this period, governments and political parties changed, as did changing international political contexts. These two discursive events indicate turning points in policy and have therefore been used to delineate the scope of the study. Other documents published before 1999 or after 2013 have been mentioned to support the discussion and offer context but are not part of the primary analysis.
2. The second criterion for inclusion was reviewed in the initial phase of the study and was expanded to include texts that impacted international students, in addition to those which had international students as their main topic. The impact here is understood as limiting or facilitating actions international students may take, from acquiring a visa, to working, studying or altering classroom practices.
65
3. The third criterion also changed to reflect the nature of policy formation, which involves a range of quasi-governmental organisations (QUANGOs) and non-departmental public bodies, as described in Chapter 1. Texts were included from any organisation which implemented, led or influenced these policies. This is consistent with applications of Bacchi’s approach incorporating documents from a range of policy actors (e.g. Lombardo and Meier, 2009; Spanger, 2011; Loutzenheiser, 2014).
4. The last criterion is that the documents be publicly available (Ashwin, et al., 2015). Because the focus is on policy discourses, internal documents such as emails, memos or information obtained through interviews would not illustrate the way that international students are talked about in public policy fora. Therefore, access to archives or privileged information was not sought. This is in keeping with Bacchi’s approach (2009; 2012) to seeking to understand how the public is governed by the discursive production and creation of problems in policy. This corpus can therefore not illustrate the mechanisms behind policy processes, the personalities or motivations involved. However, it can, through imposing these strictures, shed light on those discourses most likely to filter into everyday discourses and therefore to impact students and those who interact with them.
Because the corpus is restricted to public discourses, the ethical issues were few. With no primary informants, no informed consent needed to be sought. Anonymity was not granted to those public figures named in the primary documents, as in publishing documents or making speeches, consent was presumed. Other ethical issues have been taken into consideration here. For instance, although all information may be publicly available, it is nonetheless essential to accurately represent all statements. While it is unlikely that this research could negatively impact for any individuals named here, there is no implication of personal responsibility. Rather, they are understood to be agents within an institutional and discursive context, participating in its formation, but not personally or solely responsible for its existence. The major ethical concern in this study has been how the findings may be used to perpetuate structural inequalities, given that the thesis is a discursive event itself. The aim is to explore whether policy discourses construct images of
66
international students which may constrain their potential actions. It can only be an ethical study if it reveals, rather than perpetuates, the workings of power within discourse and policy.
Documents were identified through a combination of web searches, database searches, and use of the National Archive. Figure 14 presents the keywords used. This method resulted in small, though highly relevant, core sample of documents.
Sites Keywords
UK Government Web Archive
http://webarchive.nationalarchives.gov.uk/adv_search/ ‘international students’
‘overseas students’ ‘foreign students’
‘Prime Minister’s Initiative’ ‘International Education Strategy’
UK Government site www.gov.uk
British Council www.britishcouncil.org
Search Engines Above + UK + policy /government News sites www.bbc.co.uk www.google.co.uk/news
Figure 14: Initial text identification
In addition, references and inter-textual links were followed up, so that when one document mentioned another, I would locate and include the second document by a full title search on the above sites, or organisational home pages. For instance, the Vision 2020 report (Böhm, et al., 2004) was mentioned in multiple documents (DfES, 2004; UKCOSA, 2004; Bone, 2008, Conlon, et al., 2011), so I subsequently included it, on the grounds that it appeared to have been influential in policy discourses. Throughout the study, new documents were identified through this approach, until saturation was reached.
Classification
A range of different genres which represent and refract policy differently were included (Bacchi, 2009): press releases, speeches, research reports, evaluation and impact reports, and House of Commons minutes (see Appendix 2). Although many studies which apply Bacchi’s approach restrict their sampling to a single genre (Iverson, 2012; Svender, et al., 2012; Lancaster and Ritter, 2014), others take a more eclectic approach (Lombardo and Meier, 2009; Spanger, 2011; Zoellner, 2012;
67
Stevenson, 2013; Loutzenheiser, 2014). This study is consistent with the latter, a valid choice given the diffuse nature of the policy field as highlighted in Chapter 1. A comprehensive list of the final corpus and their provenance can be found in Appendix 2.
To create a fine-grained analysis, acknowledging the importance of provenance, documents were categorized as either primary or secondary. Documents were classified as primary if they were part of central policy initiatives, either on the basis of the position of the publishing organisation in the policy process, receipt of government funding, or the genre of the text. As such, primary documents are understood as policy. Thus, the International Education Strategy is a primary document, because it is published by a government department (provenance) and because it is a formal strategy (genre). Similarly, Tony Blair’s various speeches are considered primary because they are written by the Prime Minister (provenance), although the genre is less formal.
Secondary policy documents are those which express, reflect, or evidence policy discourses but are not published by policy making bodies. For example, House of Commons Select Committee evidence and reports are included in the corpus because they offer a window into public policy discourses, and are classified as secondary because they are not central to the policy process. The Select Committees can publish findings and recommendations, but the Government is not obligated to abide by them. Similarly, research reports conducted on behalf of UKBA or BIS may reflect and inform policy discourses, but do not necessarily determine policies themselves.
In certain cases, funding and governance changes during the period (see Chapter 1) mean that provenance from the same organisation leads to different categorisation in different years. For example, certain British Council documents were classified as primary where they related to implementing PMI and PMI2 because they led the initiative and received Cabinet Office funding to do so. After the PMI2 concluded in 2011, British Council reports were classified as secondary because they were no longer central to the policy process, although they are often influential. A list of the
68
institutional policy actors and their responsibilities is included in Appendix 1, and Appendix 2 lists full document details with provenance and categorisation.
While greater attention has typically been given to the primary policy documents, differences and contradictions from the main narrative have been highlighted (Kuckartz, 2014) to illustrate the discursive complexity of the field. Because the research aimed to capture public policy discourses, voices from the HE sector and student organisations are incorporated in discussion and analysis, but were excluded from the data collection.
Procedures
The analytical procedure followed is depicted visually in Figure 15 and explained in more detail below.
Document collection First reading (manual, on paper) Initial open coding (CAQDAS, digital)
Second order coding (merging, renaming, creating hierarchy) Selective coding (identifying the relevant concepts)
Problematisation analysis (Bacchi’s 6 questions)
Figure 15: Stages of the research
First, I read the texts in detail to identify the sections relevant to international students. The purpose of this initial reading was to establish a chronology of policy discourse. It also raised my awareness of key themes for initial open coding.
Then I coded the documents inductively, based on emerging themes. Inductive thematic coding is a common choice for qualitative data analysis, as it allows the researcher to approach the data with fewer preconceptions than a deductive framework (Braun and Clark, 2006). In essence, inductive or open coding involves assigning codes as they emerge from the data, and is ‘data driven’ rather than theoretically driven (Gibbs, 2008). Not all codes from this stage were retained in the final analysis, such as ‘Other sectors’, which included reference to English language teaching and TNHE. While deductive coding has been used in combination with Bacchi’s framework (e.g. Iverson, 2012), initial experimentation suggested that this
69
would be unfruitful for this study. These policy documents often keep the problem implicit, emphasising solutions instead. Problems and silences must therefore sometimes be inferred from their absence (Bacchi, 2009) and it is not always possible to identify them in the text through a deductive coding system.
A Computer Assisted Qualitative Data Analysis Software (CAQDAS) programme, NVivo, facilitated the analysis. NVivo facilitates developing a hierarchy of codes, allowing certain codes to become ‘parent nodes’, main themes, and others ‘child nodes’, which retains differentiation within the hierarchy. It allows qualitative researchers to establish an electronic filing system, by tagging portions of text and assigning them a code. Reports can then be produced which compile all extracts with the same label. Its main utility is therefore to facilitate the retrieval of information attached to codes. It also makes simultaneous coding easier. Another essential function was text searching, used to supplement manual coding. These advantages enabled a large volume of documents to be included in the corpus: 103 documents in total were included, with over 3,000 pages. NVivo also helped to maintain consistency in coding because previous code reports can be quickly checked and changed without difficulty.
A research diary was kept throughout (see Figure 16).
70
I kept notes of observations made during the coding process, and reminders to myself of additional documents to follow up and cross-checks to do later. Evernote (a web application) was used to record analytical insights (Bazeley and Jackson, 2013). These were organised into notebooks and tagged with keywords. Relevant entries were then retrieved during analysis either by tags or by word searching. This research diary also allowed me to track where I stopped work on a given day and what I had left to do. Example entries are included in Appendix 4.
Coding
The documents were coded using an open, inductive approach in two principal stages.
Initial coding
In the first stage of coding, after a preliminary reading for familiarisation, extracts were coded if they related to the research questions, like Ashwin et al. (2015). Sentences which reiterated background already coded from the introduction were excluded, for instance, as were background sentences which did not directly relate to international students, and research report methodologies. Sentences relevant to the research questions were established by examining regularities, disjunctures, and attention to discourse (Bazeley and Jackson, 2013). For example, in the following extract, the first two sentences were coded (simultaneously to ‘satisfaction’, ‘experience’ and ‘teaching’), where the last (underlined) was not:
Students’ academic experiences depend largely on good teaching. A new project, Teaching International Students, funded jointly by PMI2 (through UKCISA) and the Higher Education Academy...
For more information on all the projects listed above, please go to www.ukcisa.org.uk/pmi www.studentcalculator.org.uk/international
www.prepareforsuccess.org.uk. (BC, 2010, p.21).
The underlined information refers readers to the websites for more information and does not contain any information pertinent to the research questions. This demonstrates the type of information that was not coded at all.
Open coding began from the primary policy documents, then proceeded to the
secondary policy documents, to establish codes from the most central documents. Within each category (primary and secondary), I worked in chronological order to
71
gain a sense of changes over time, enabling constant comparison (Gibbs, 2008). Codes names were assigned descriptively or In Vivo (Saldana, 2009). I coded ‘close to the documents’, using open coding and labelling the sections of text with language and terminology derived from the documents themselves (Strauss and Corbin, 1990; Bazeley and Jackson, 2013). Many codes became more descriptive as the coding structure evolved, starting off with a phrase taken directly from the documents, and gradually being modified to become more widely applicable. However, certain codes were retained with their original In Vivo phrasing where this was particularly relevant, such as ‘the brightest and the best’, as this phrase was frequently used. Where this differentiation was thought significant, the original code was retained as a child node under a broader parent node.
Inductive coding meant that when new codes were added, previously coded
documents needed to be reviewed for relevance to these codes. I used text searches to find these. This enabled the constant comparison that Gibbs (2008) argues is essential to good qualitative analysis; it maintained consistency and ensured that the same codes were used throughout the study. In Nvivo, text searches can be produced in either ‘narrow’ (see Figure 17, for example) or ‘broad’ context (see Figure 18).
Figure 17: ‘Narrow view’ of a text search for ‘teaching’
The ‘narrow context’ was often found to be inadequate to determine relevance. Instead the ‘broad context view’ was used (Bazeley and Richards, 2000), to decide whether the phrase matched the parameters of the new code.
72
For example, in the case of the ‘teaching’ code, I set the parameters as ‘teaching international students’. Broad context results presented in Figure 18 emerged from Chapter 4 of the White Paper The Future of Higher Education (DfES, 2003) and were generic to all students and, therefore, matches were not coded to the ‘teaching’ node.
Figure 18: ‘Broad view’ of a text search ‘teaching’
Where the broad context proved inadequate, NVivo permitted the navigation directly to the result in situ in the original document through the hyperlink in the report. This is a positive argument for the use of CAQDAS in contradiction of early critiques that such analysis removes text from its context (Bazeley and Jackson, 2013).
Automatic coding was occasionally used to identify relevant key phrases, such as
‘the international student experience’ and ‘the best and the brightest’, which were used frequently verbatim. Automatic coding was used to ensure that all instances were coded. Reports were then checked to remove irrelevant results. It was not used otherwise, in order to preserve the integrity of an inductive approach.
The unit of coding varied according to the document, either line-by-line or a whole paragraph, depending on the document (Strauss and Corbin, 1990; Saldana, 2009). Line-by-line coding occurred in the thematically dense areas, which were also often simultaneously coded where a particular sentence incorporated several themes. This lent a sense of the richness of the data (Bazeley and Jackson, 2013). It was more common to code by sentence units because the main body of most texts usually dealt
73
with a single issue or theme at a time. In some texts, the same code was applied to several sequential sentences or whole paragraphs, in what Bazeley and Jackson (2013) call a ‘lumper’ approach, a broad-brush sorting. For instance, in coding the Home Affairs Select Committee report on bogus colleges, the majority of the document was coded to the single code ‘bogus colleges’ because although a number of more specific issues were raised in this document, these were not pertinent to the research questions. Bazeley and Jackson (2013) suggest that this should usually be a first-order approach to coding, and a finer-grained analysis is certainly possible of these excerpts. However, for the purposes of this study, ‘bogus colleges’ was thought to be a sufficiently nuanced code in reference to the research aims of this study.
Second order coding and revisions
‘Second order’ or ‘axial coding’ involved reviewing each code individually for consistency. It also established relationships and connections between codes expressed through a hierarchy, reviewing redundant nodes, and renaming codes for consistency (Gibbs, 2008). During this review, codes were merged, and hierarchies of parent and child nodes were further developed (Saldana, 2009; Bazeley and Jackson, 2013).
Codes were merged when very similar meanings or patterns of coding were
apparent. Simultaneous coding was reduced to occur only when two nodes which are organised under separate parent codes, such as ‘export earnings’ and ‘student visa system’ can both be applied to the same statement. Compound queries were used to check the comprehensiveness of coding. For example, I checked whether all mentions of ‘visas’ in the primary policy documents had indeed been coded to the ‘migration’ parent code by excluding any that matched both criteria.
Internal consistency of codes was reviewed by exporting the reports and checking
them manually. Given that I was the only coder, Bazeley and Jackson (2013) suggest